Toward user-driven Metaverse applications with fast wireless connectivity and tremendous computing demand through future 6G infrastructures, we propose a Brain-Computer Interface (BCI) enabled framework that paves the way for the creation of intelligent human-like avatars. Our approach takes a first step toward the Metaverse systems in which the digital avatars are envisioned to be more intelligent by collecting and analyzing brain signals through cellular networks. In our proposed system, Metaverse users experience Metaverse applications while sending their brain signals via uplink wireless channels in order to create intelligent human-like avatars at the base station. As such, the digital avatars can not only give useful recommendations for the users but also enable the system to create user-driven applications. Our proposed framework involves a mixed decision-making and classification problem in which the base station has to allocate its computing and radio resources to the users and classify the brain signals of users in an efficient manner. To this end, we propose a hybrid training algorithm that utilizes recent advances in deep reinforcement learning to address the problem. Specifically, our hybrid training algorithm contains three deep neural networks cooperating with each other to enable better realization of the mixed decision-making and classification problem. Simulation results show that our proposed framework can jointly address resource allocation for the system and classify brain signals of the users with highly accurate predictions.

翻译：面向未来6G基础设施中具有快速无线连接和巨大计算需求的用户驱动型元宇宙应用，我们提出了一种基于脑机接口（BCI）的框架，为创建智能类人虚拟形象铺平道路。该方法迈出了元宇宙系统的第一步，其中数字虚拟形象通过蜂窝网络收集并分析脑信号，被设想为更具智能化。在我们提出的系统中，元宇宙用户在使用元宇宙应用的同时，通过上行无线信道发送脑信号，以便在基站端创建智能类人虚拟形象。由此，数字虚拟形象不仅能为用户提供有益建议，还能使系统创建用户驱动的应用。该框架涉及一个混合决策与分类问题，即基站必须高效地向用户分配计算和无线资源，并对用户脑信号进行分类。为此，我们提出一种混合训练算法，利用深度强化学习的最新进展来应对该问题。具体而言，该混合训练算法包含三个深度神经网络相互协作，以更好地实现混合决策与分类问题的求解。仿真结果表明，所提框架能够联合解决系统的资源分配问题，并以高度准确的预测对用户脑信号进行分类。